Nonprofit Sector and Part-Time Work: An Analysis of Employer-Employee Matched Data of Child Care Workers

Transcription

1 DISCUSSION PAPER SERIES IZA DP No. 408 Nonprofit Sector and Part-Time Work: An Analysis of Employer-Employee Matched Data of Child Care Workers H. Naci Mocan Erdal Tekin December 2001 Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor

2 Nonprofit Sector and Part-Time Work: An Analysis of Employer-Employee Matched Data of Child Care Workers H. Naci Mocan University of Colorado at Denver and NBER Erdal Tekin Andrew Young School of Policy Studies, Georgia State University and IZA, Bonn Discussion Paper No. 408 December 2001 IZA P.O. Box 7240 D Bonn Germany Tel.: Fax: This Discussion Paper is issued within the framework of IZA s research area Mobility and Fexibility of Labor. Any opinions expressed here are those of the author(s) and not those of the institute. Research disseminated by IZA may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent, nonprofit limited liability company (Gesellschaft mit beschränkter Haftung) supported by the Deutsche Post AG. The center is associated with the University of Bonn and offers a stimulating research environment through its research networks, research support, and visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. The current research program deals with (1) mobility and flexibility of labor, (2) internationalization of labor markets, (3) the welfare state and labor markets, (4) labor markets in transition countries, (5) the future of labor, (6) evaluation of labor market policies and projects and (7) general labor economics. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character.

3 IZA Discussion Paper No. 408 December 2001 ABSTRACT Nonprofit Sector and Part-Time Work: An Analysis of Employer-Employee Matched Data of Child Care Workers * This paper uses a rich employer-employee matched data set to investigate the existence and the extent of nonprofit and part-time wage and compensation differentials in child care. The empirical strategy adjusts for workers self-selection into the for-profit or nonprofit sectors, into full-time or part-time work, as well as for unobserved worker heterogeneity using a discrete factor model. We find differences between the regimes (full-time for-profit, full-time nonprofit, part-time for-profit, part-time nonprofit) in the manner in which human capital characteristics of the workers are rewarded. There is substantial variation in wages as a function of employee characteristics, and there is variation in wages within sectors. The results indicate that part-time jobs are good jobs in center-based child care, and there exist nonprofit wage and compensation premiums, which support the property rights hypothesis. JEL Classification: J2, J3, J5, L3 Keywords: Nonprofit sector, employment, child care Erdal Tekin Department of Economics Andrew Young School of Policy Studies Georgia State University University Plaza Atlanta, GA USA Tel.: +1 (404) Fax: +1 (404) tekin@gsu.edu We thank Damba Lkhagvasuren, Paul Niemann and Kaj Gittings for excellent research assistance, Brian Duncan for providing the Census data and an anonymous referee for helpful suggestions. Mocan gratefully acknowledges financial support through an NBER Nonprofit Sector Research Grant.

4 Nonprofit Sector and Part-Time Work: An Analysis of Employer-Employee Matched Data of Child Care Workers I. Introduction The nonprofit sector constitutes a significant, and expanding segment of the U.S. economy. The number of private nonprofit organizations increased by almost 19 percent between 1989 and 1997, from 1,262,000 to 1,498,000 (Urban Institute 2000). The number of national nonprofit associations increased by 56 percent between 1980 and 1997, reaching almost 23,000 in 1997 (Statistical Abstract of the U.S. 2000). Employment in the nonprofit sector increased from 7.1 million full-time equivalent (FTE) paid workers in 1990 to 9.6 million FTE workers in 1995; and in 1995 nonprofit employment accounted for 8.8 percent of the GDP in the U.S. (Salamon et al. 1999). There are theoretical reasons to believe that economic behavior of nonprofit enterprises may be different from their for-profit counterparts (Lakdawalla and Philipson 1998, Rose-Ackerman 1996, Hansmann 1980). One such difference pertains to wage-setting behavior between nonprofit and for-profit firms. Nonprofit enterprises are expected to create rents for their workers, which would translate into wage mark-ups. Alternatively, if nonprofit workers differ from for-profit workers in their altruism, this may translate into a nonprofit labor donation and lower nonprofit wages (Preston 1989). The empirical evidence on nonprofit wage differentials is ambiguous. As explained in Section II below, the literature is far from having reached a consensus on the issue. Around 17 percent of all workers, and around one-quarter of all female workers work part-time (Bureau of Labor Statistics 2000). The fact that average wages of part-time workers are less than those of full-time workers with similar characteristics raises the question of whether part-time jobs are bad jobs (Blank 1990). 1 The substantial size of part-time employment makes this an important, and as yet unresolved, question. Using an extraordinarily detailed employer-employee matched data set, this paper addresses two questions: Are wages and compensation of workers who work in the nonprofit sector lower than their counterparts in the for-profit sector; and are part-time jobs bad jobs? 1 An investigation of whether certain jobs are bad jobs requires data on worker compensation as well as controls for job characteristics, which is not always possible because of the paucity of data. This 1

5 We find that both answers are no for child care workers. We document substantial nonprofit premiums in wages and compensation. Similarly, we find that part-time jobs are good jobs. These results are not likely due to some idiosyncratic characteristic of the child care industry. As we demonstrate in the paper, the raw full-time and nonprofit differentials observed in our data show similar patterns when compared to a number of industries of the 1990 census data. Section II gives the background, and puts the contribution of this paper into perspective. Section III presents the model, section IV describes the data and section V presents the results. Section VI includes the sensitivity analysis and section VII is the conclusion. II. Background As summarized by Preston (1988), the theory of property rights predicts that nonprofit institutions may generate rents for their workers. This is partly because there is no owner to which the nonprofit manager is held accountable, which lessens the incentive for managers of nonprofit organizations to operate efficiently. As a result, there is reason to expect that nonprofit enterprises pay higher wages to comparable workers than their for-profit counterparts. On the other hand, it can be argued that nonprofit workers may have preferences that are different from for-profit workers. More precisely, some workers may be willing to work at nonprofit institutions at a lower wage and/or compensation in comparison to wages and compensation they could have obtained elsewhere. This labor donation could take place if nonprofit workers care about the social value of the good or service they produce more than for-profit workers (Preston 1989). However, the empirical evidence on the nonprofit wage differential is ambiguous. Most of this ambiguity seems to stem from inadequate data sets employed to address the question. The investigation of nonprofit wage differentials takes two general forms. The first one is the analysis of firm-level data, such as in Preston (1988), who found that in the competitive segment of the day care industry, there is no significant difference in firm-level salaries between nonprofit and for-profit firms, whereas there is a 5 to 10 percent nonprofit differential in the government subsidized segment of the industry. Mocan and Viola (1997) issue is discussed below in more detail. 2

6 investigated the determinants of wages and compensation in 398 child care centers. They found a positive nonprofit premium, which became statistically insignificant once sector breakdowns (e.g. religious center, publicly funded center) are controlled for. A more refined investigation is done with micro data, where the determinants of worker wages are analyzed as a function of, among other factors, nonprofit status. Some analysts estimated wage regressions by including a dummy variable to indicate sector affiliation of the worker (e.g. Shackett and Trapani 1987, Borjas, Frech III and Ginsburg 1983) The findings of these studies may not be reliable because of the potential endogeneity of the sector dummy. As argued above, it is conceivable that workers who choose to work in the nonprofit sector may be systematically different from workers who work in the for-profit sector. If unobservable worker characteristics which influence sector choice also impact workers wages, one would obtain biased estimates of the sector impact. For example, Weisbrod (1983) estimated separate earnings functions for lawyers employed in public sector firms and those employed in private firms. He reported that public sector lawyers received lower wages than private sector lawyers. Using the same data set and by controlling for self-selection into sectors Goddeeris (1988) found no evidence that public sector lawyers accepted earning sacrifices, underlining the importance of selection bias. There are only a few papers that addressed selfselection of workers into sectors. Holtmann and Idson (1993) and Preston (1989) used the two-stage approach developed by Lee (1978) and Heckman (1979). Ruhm and Borkoski (2000) exploited the panel component of the Current Population Survey Outgoing Rotation Groups to control for self-selection. For the most part, recent research on part-time wage differentials has paid attention to non-random selection into part-time work (Hotchkiss 1991, Blank 1990, Simpson 1986). However, most of the work in this literature faced challenges pertaining to measurement error in key variables (such as wages and hours), and a failure to control for non-wage benefits and other job attributes. As explained by Montgomery and Cosgrove (1995), the data sets employed in these studies did not allow for detailed controls for firm effects, or even occupation effects. While most papers reported a negative part-time wage differential (Montgomery and Cosgrove 1995, Simpson 1986), Blank (1990) found a wage differential in favor of part-time workers. This paper brings together these two strands of literature: the wage setting practices of 3

7 nonprofit enterprises, and the investigation of relative wages of part-time workers in comparison to their full-time counterparts. The employer-employee matched data set used in the analysis not only includes very detailed characteristics of workers and firms, it also allows for measurement of variables with more precision than before. For example, most studies that used national data sets had to impute worker wages using annual wage or salary income and annual hours (Leete 2001, Main and Reilly 1992, Blank 1990). They were also forced to impute certain human capital measures of the workers, such as experience. Similarly, data obtained from workers are likely to contain substantial error regarding the nonprofit status of the establishment they are working for (Leete 2001, Ruhm and Borkoski 2000). By contrast, in our data set the information pertaining to the firm is obtained from child care centers directors and from their financial statements. All worker information such as workers experience in the field, experience outside the field, and tenure on the job, is obtained by surveying the workers directly. Using the information on the types and dollar amount of nonwage benefits offered by the centers, and the information on the type of staff which receives these benefits, we are able to calculate hourly compensation, and conduct the analysis for compensation as well. We allow for endogenous selection into sectors (nonprofit vs. for-profit) as well as hours (part-time versus full-time). Controlling for unobserved individual heterogeneity, the model is estimated for both wages and compensation using full-information maximum-likelihood. III. Empirical Implementation The model includes two selection equations and four sectoral wage equations. One selection equation allocates child care workers to the for-profit or the nonprofit sector, while the other one allocates them to either full-time or part-time work. The cross-classification of these two selection rules partitions workers into four mutually exclusive categories. 2 The selection between the for-profit and nonprofit sectors and between full-time and part-time work can be summarized by the following equations. (1) PR * i=xi+ i pr 2 Another potential selection is into the labor force. We do not address it here because of the lack of data on nonworkers. 4

8 (2) FT * i=yi+ i ft where i represents workers. PR * stands for unobserved sentiment that determines the attachment to the for-profit sector, and FT * is the unobserved tendency to choose full-time work, such that PR=1 (the worker chooses the for-profit sector) if PR * >0, and PR=0 otherwise. Similarly, if FT * >0 then the worker chooses full-time work and the dichotomous variable FT takes the value of 1, and FT=0 otherwise. The vectors X and Y are variables that determine the propensity to work in the for-profit sector, and to work full-time, respectively. We specify linear sectoral wage equations as follows. (3) Wi ft-np =Zi + i ft-np (4) Wi pt-np =Zi + i pt-np (5) Wi ft-pr =Ki*+ i ft-pr (6) Wi pt-pr =Ki&+ i pt-pr, where the superscript ft stands for full-time, pt stands for part-time, np is nonprofit and pr represents for-profit. Wi stands for the wage rate of the ith worker. The vector Z contains variables determining the wages in the nonprofit sector, and K is the vector of variables that determines wages in the for-profit sector. Although the model is identified through nonlinearities, to further facilitate identification the variables that represent family background such as household income, marital status of the worker, and the number of children are included in the selection equations, but they are omitted from the wage equations. To the extent that the latent variables that measure the propensity to work full-time and the propensity to work in the for-profit sector are influenced by potential wages in these sectors, the variables that are included in Z and K should be included in X and Y. However, variables that are affiliated exclusively with one sector (such as the union status) are not included in the selection equations. It is plausible to think that the error term of the sector selection equation may be correlated with the error term of the full-time selection equation. That is, workers unobserved preferences for sector choice may impact their choice of full-time versus part-time work. These disturbances may also be correlated with the disturbances of the wage equations: unobserved worker characteristics which influence sector and full-time choice decisions may impact wages. 5

9 To account for this potential correlation in the errors of the equations, we model the error structure as 2 (7) i pr =ui pr + /,j pr j j=1 2 (8) i ft =ui ft + /,j ft j j=1 2 (9) i ft-np =ui ft-np + /,j ft-np j j=1 2 (10) i pt-np =ui pt-np + /,j pt-np j j=1 2 (11) i ft-pr =ui ft-pr + /,j ft-pr j j=1 2 (12) i pt-pr =ui pt-pr + /,j pt-pr j j=1 where ui pr,, ui pt-pr are mutually independent disturbances with mean zero, and s symbolize the common factors that impacts error terms. The factor loadings,, s, allow for the impact of the common factors to vary cross equations. u s and s are independent of the explanatory variables. The factors ( s) are unobserved variables that generate worker heterogeneity. We allow for two factors to account for the correlations among equations. For example, one factor might represent unobserved motivation, and the other one may represent tastes for work. In principle, one can estimate this system by imposing a parametric joint distribution for the two sources of unobserved heterogeneity and integrating out over its distribution. The drawback to this approach is that it requires computing multi-dimensional integrals, which is not feasible with traditional methods. Furthermore, it requires strong assumptions about the exact distribution of the heterogeneity. In related contexts, it has been shown that the results can be sensitive to departures from multi-normality (Goldberger 1983, Arabmazar and Schmidt 1982). In this paper we use the discrete factor method (DFM) which, unlike standard selection corrections, estimates a semiparametric distribution to approximate the distribution among the error terms of the selection and wage equations. In this approach, the distribution of the ν s is approximated with a step function and integrated out through a weighted sum of 6

10 probabilities (Heckman and Singer 1984). Following Mroz (1999), we assume that is governed by a discrete distribution (13) Prob( =µk)=4k; k=1,..., K; 4k 0,!k4k=1, where µk are the points of support of the distribution, and 4k are the probability weights. The µk s, 4k s, and, s are parameters to be estimated. K is specified a priori, and the six equations are estimated jointly with full-information maximum likelihood. 3 Mroz (1999) demonstrates that when the true correlation of the error terms is multi-normal, DFM performs well in comparison to estimators which assume multi-normality; and DFM performs better than normality-based estimators when the underlying distribution is non-normal. (See Blau and Hagy (1998), Hu (1999), Blau and Mocan (1999) and Mocan, Tekin and Zax (2000) for applications of the discrete factor model). IV. The Data We use a data set obtained from child care centers in California, Colorado, Connecticut and North Carolina. 4 The data are based upon a stratified random sample of 398 day care centers (approximately 100 centers from each participating state), with equal representation of for-profit and nonprofit programs, providing full-time year-round care. They were obtained by actual visits to the centers during the spring of Data collectors gathered in-depth financial information on center costs and revenues, various non-wage benefits offered to teachers, aides and part-time staff, ownership and union status, and a host of other center characteristics through on-site interviews and reviews of center records with center administrators or owners. In each of the 398 centers in the data set, two classrooms were randomly selected. Teaching staff in these classrooms were asked to complete a survey regarding their human capital characteristics, their pre-tax hourly wages, weekly hours of work, family characteristics and attitudes towards work in the child care industry. As a result, the extraordinary detail of the data allows for control of the job environment, as well as worker and firm characteristics with precision. 3 The likelihood function is presented in the Appendix. 4 The data were compiled with the collaboration of economists, psychologists and child development experts from University of Colorado at Denver, Yale University, University of North Carolina at Chapel Hill and UCLA. The details of the data and data collection can be found in Mocan (1997). 7

11 Each center provided information about the total dollar value of their nonwage benefits. Each centers also provided detailed information about 11 different benefits provided for three worker categories (teacher, assistant teachers, and part-timer) that involve monetary costs to the center. For example, centers indicated whether fully-paid or partially-paid health insurance is provided for teachers, assistant teachers and part-timers as a benefit. Other examples are at least partially paid dental insurance, paid vacations, paid maternity leave, paid health insurance for dependents, and paid sick or personal leave. For each center the total number of benefits provided for each worker category is counted. Because the job title and hours of work for each worker is known, the annual number of hours worked by full-time teachers, annual hours of full-time aides and annual part-time hours can be calculated. Using this information, and center s total spending on benefits, as well as information on the proportion of line-item benefits provided to each worker group, the monetary value of hourly benefits are calculated that can be assigned to each group of worker. Hourly compensation for each worker is her reported hourly wage plus hourly non-wage benefits for her job category. Raw full-time and nonprofit differentials are presented in Table 1. We ran linear regressions of the logarithm of wages and compensation on state dummies and a full-time dummy, separately for nonprofit and for-profit centers to obtain the information displayed in the top half of Table 1. The logarithm of wages and compensation are run on state dummies and a for-profit dummy separately for full-time and part-time workers to obtain the information displayed in the bottom half of the table. The values in parentheses are percentage differences. According to Table 1, controlling for state differences, full-time wages are not different from part-time wages in nonprofit centers, but full-time compensation is greater than part-time compensation. In for-profit centers both wages and compensation are greater for full-time workers. The bottom half of the Table demonstrates that there exist nonprofit mark-ups in both wages and compensation. Table 2 presents worker characteristics by type of worker. A star indicates that the means are statistically different between for-profit and nonprofit, or between full-time and part-time workers at the five-percent level or better. For example, nonprofit workers are older than forprofit workers (35 years of age vs. 31 years), and they also have more tenure (45 months vs. 30 months). The workers were asked about the main reason for their choice of work in the field of early education and child care. The alternatives given were: it was the highest paying job at the 8

12 time; it is a desirable job (pleasant job environment); low cost of working; flexible working hours; and this is an important job that someone needs to do. If the worker chose this last alternative, then the dummy variable Important job takes the value of one; and zero otherwise. This variable is a direct indicator of the intrinsic value to the individual of working in the child care sector. As such, it allows for a direct test of the labor donation hypothesis; that is, all else the same, individuals with this attitude are expected to command lower wages, especially in the nonprofit sector. Table 3 displays the descriptive statistics of center characteristics. Publicly Regulated is 1 if the center receives public money, either from the state or federal government, tied to higher standards (above and beyond normal licensing regulations), and 0 otherwise. This group includes Head Start programs, centers where 20 percent or more of their enrollment constitute special needs children, special preschool programs sponsored by the State or Federal Department of Education, and other special programs in Connecticut and California. Publicly owned is set to 1 for centers that are owned and operated by public agencies. Examples include public colleges, hospitals, and city departments of family services. Religious is 1 if the center has a religious affiliation (e.g. church-based centers), and zero otherwise. Union is 1, if teachers and/or aides are unionized. There are 18 unionized centers in the sample, and all are nonprofit centers. Of these, 14 are publicly owned centers (Publicly owned), 2 are publicly supported centers (Publicly supported) and 2 are religious centers. Percent subsidized represents the proportion of children that are subsidized by a government or other agency, such as the State or County Department of Social or Human Services, United Way, etc. Table 3 also displays information about various programs offered by centers, such as evening care, sick care, and before-and-school care. Following an extensive literature that investigates the impact of firm size on earnings (Troske 1999, Main and Reilly 1993, Brown and Medoff 1989, Evans and Leighton 1989), we add the size of the child care center, measured by full-time equivalent children (FTE Children) in the wage and compensation equations. Local unemployment can depress wages through various non-competitive mechanisms (Blanchflower and Oswald 1994, Chapter 3). Alternatively, there can be a positive relationship between wages and unemployment, because firms can provide compensating wage differentials to workers in the spirit of the Harris and Todero (1970) and Hall (1970) models. To investigate this effect, we include the 9

13 unemployment rate in the models. Unemployment is the unemployment rate in the city where the center is located. The zip codes of the centers are used to determine the city in which the center is located. The unemployment rates for the corresponding cities are obtained from the U.S. Bureau of Labor Statistics, Local Area Unemployment Division. For the cities where BLS figures were unavailable, the unemployment rates are imputed as the average unemployment rates of the neighboring cities within a 20 mile radius. V. Results Equations (1)-(6) are estimated jointly with full-information maximum likelihood under the error structure displayed by (7)-(12). The results are based on five points of support (K=5 in Equation 13). The discussion of the estimated coefficients of the selection equations are not reported, but can be found in Mocan and Tekin (2000). 5 Table 4 displays the estimated coefficients of the four wage equations. The dependent variables are the logarithm of wages. We observe that female part-time workers earn 11 percent and four percent less, respectively, in the for-profit and nonprofit sectors in comparison to their male counterparts. Race is also a determinant of wages, where blacks and Hispanics receive positive premiums, and white and Asian part-time workers in the nonprofit sector command lower wages. Table 5 presents the results of a series of hypothesis tests pertaining to the equality of the coefficients of human capital variables. 6 The sign of the calculated t-statistic reveals the direction of the difference between the coefficients under test. A positive t-statistic indicates that the first coefficient under the hypothesis is greater than the second one in magnitude. 7 example, the first test reported in the table pertains to the hypothesis that the coefficients of tenure are equal to each other in for-profit and nonprofit sectors for full-time workers. The calculated t-value is 3.627, which indicates that the coefficients are different from each other, and the for-profit coefficient is larger. As Table 5 demonstrates there are differences between for-profit and nonprofit centers For 5 These tables can also be downloaded from or 6 The hypothesis tests utilize the information reported in Table 4 along with the variance-covariance matrix of the estimated coefficients. 7 The exception is the test pertaining to the coefficients of Special Training. Because these coefficients are always negative, a positive t-statistic indicates that the first coefficient is smaller than the second one in absolute value. 10

14 regarding the manner in which human capital characteristics affect wages. For full-time workers, the return to tenure is higher in for-profit centers, but the return to experience, return to a college degree and to a graduate degree are higher in nonprofit centers. Section II shows that experience in the child care sector, other experience, and a graduate degree are rewarded more heavily in the nonprofit sector for part-time workers. Similarly, the lack of special training in child care lowers wages of part-time workers more in the nonprofit sector. On the other hand, a community college degree has a higher return in the for-profit sector. With the exception of Other Experience in Section III and Experience in Section IV, all significant t- values have negative signs in Sections III and IV of Table 5. This demonstrates that part-time coefficients are larger in magnitude than full-time coefficients in both for-profit (Section III) and nonprofit centers (Section IV), and indicates that the return to education is higher for parttime workers in both sectors. To the extent that the variable important job is capturing the attitudes toward labor donation of nonprofit workers, its coefficient should be negative in the nonprofit wage equation. Table 4 shows that the coefficient of important job is negative for both full-time and part-time workers in the nonprofit sector, with a bigger magnitude for full-time workers. It is positive in the for-profit sector, although statistically significant only for full-time for-profit workers. This is an interesting result, which provides support to the labor donation hypothesis for nonprofit workers. It also suggests that this particular attitude may be correlated with productivity, which is rewarded in the for-profit sector. 8 Center characteristics have significant impacts on wages. For example, wages in onsite centers are substantially higher for both full-time and part-time workers in both sectors. For-profit centers, which are part of a national chain, offer wages to full-timers and part-timers that are six percent and 12 percent lower, respectively, in comparison to non-chain for-profit centers. Wages are three-to-four percent lower in religious centers. There is a substantial union wage premium in nonprofit centers. Unionized nonprofit centers provide a 22 percent wage premium to both full-time and part-time workers. The union variable in the analysis indicates whether or not the center is unionized. The impact of this union variable on part-time wages has two possible explanations. The first explanation is 8 A more detailed discussion of the results on personal characteristics can be found in Mocan and Tekin (2000). 11

15 that part-time workers are also covered by the union. Alternatively, part-time workers are not covered by the union, but the union impact on part-time wages reflects a spillover effect. 9 Local unemployment has a negative impact on part-time nonprofit workers wages. Larger forprofit centers pay a wage premium to both full-time and part-time workers which is consistent with the literature on size differentials (see Troske 1999). On the other hand, there is a negative relationship between center size and wages in the nonprofit sector. In Table 6 we present the estimated coefficients compensation equations. 10 The results are similar to the ones obtained from wage regressions. To understand the extent to which the observed wage and compensation differences are attributable to observable characteristics, we estimated wage and compensation equations identical to those reported in Tables 4 and 6 using OLS. These OLS regressions attempt to explain the variation in wages and compensation within sectors without regard to selection. The results, which are reported in rows (2) of Table 7, demonstrate that, for the most part, controlling for human capital and firm characteristics reduces the size of the observed wage and compensation gaps. Our estimated model (Equations 1-13), enables us to investigate what a randomly chosen individual s wage and compensation would be under the four possible regimes (fulltime, for-profit; full-time, nonprofit; part-time, for-profit; and part-time, nonprofit). Using the estimated parameters of the system and the estimated heterogeneity coefficients, we simulated the potential wages and compensation for all workers in the sample under the four possible regimes. That is, we calculated selection-corrected and heterogeneity-adjusted wages and compensation that each worker would earn if assigned to a particular regime. The prediction errors of wages and compensation are also calculated for each worker under each regime. This allows for the calculation of the standard errors of the means of predicted wage and 9 We attempted to contact the unionized centers in the data set to ask whether part-time workers were also covered by the union in We could not reach four out of the 12 centers either because they were not in operation any longer, or their phone numbers have changed. Four of the remaining eight centers indicated that part-time workers were also covered by the union. One center indicated that part-timers were not covered by the union, and three centers could not answer the question. So, the channel through which the union impact on part-time wages operates remains unclear. 10 The results of the selection equations are reported in Mocan and Tekin (2000). They can also be obtained from or 12

16 compensation. Calculation of the 95-percent confidence intervals around the means indicate that we can reject the hypotheses of the equality of wages and compensation with the exception of full-time and part-time wages in the for-profit sector as their confidence intervals overlap. 11 Rows (3) of Table 7 present the direction and the magnitudes of these mark-ups. After controlling for selection effects and unobserved worker heterogeneity, full-time wages are 12.5 percent lower than part-time wages in nonprofit centers, and they are equal to part-time wages in for-profit centers. This is not a common result in the literature, but it is not without precedent. Blank (1990) also found that selection-adjusted wages of part-time workers were higher than comparable full-time workers. She was unable to do her analysis using compensation because of lack of data. In nonprofit centers, the compensation mark-up is also in favor of part-timers, and larger than wage mark-up in magnitude (24.9 percent). This indicates that not only part-time workers wages are higher than full-time workers wages in nonprofit centers, but part-time workers receive more benefits per hour worked than full-time workers in the nonprofit sector. Although part-time workers wages are equivalent to full-time workers wages in forprofit centers, the average compensation of part-time workers is 23 percent higher than that of full-time workers in for-profit centers, indicating that, similar to the pattern in nonprofit centers, for-profit centers provide more benefits to part-time workers. These results are in sharp contrast to the raw data revealed in Table 1, which are also displayed in rows(1) of Table 7. Row (3) of the bottom panel of Table 7 demonstrates that full-time workers wages are 5.8 percent higher in the nonprofit sector in comparison to what they would have earned in the for-profit sector. The nonprofit full-time mark-up is 7.8 percent in compensation. Part-time workers wages and compensation are higher in the nonprofit sector in comparison to the forprofit sector. However, the mark-up is smaller in compensation, suggesting that for-profit centers provide better benefits than nonprofit centers in the case of part-time workers. We also calculated selectivity corrected mark-ups, assuming no individual heterogenity, which are reported in rows (4). The mark-ups are somewhat larger in absolute value for part- 11 Var(Σ i /n)=(1/n 2 )ΣVar( i ), where i stands for the predicted wage or compensation. Var( i )=σ 2 +X(X X) -1 X σ 2, and σ 2 is the variance of the errors. The estimated standard errors for wage equations in pr-ft, np-ft, pr-pt and np-pt are , , and , respectively. They are , , and for compensation equations. 13

17 time workers wages, and smaller in case of nonprofit differentials. These results indicate that part-time jobs are good jobs in child care. The observed wage differences between full-time and part-time workers is not explained by differences in workers human capital or by differences in firms characteristics. The fact that the raw wage differences are in favor of full-time workers implies that, as suggested by Blank (1990), parttime workers may have earned lower wages even when they worked full-time, due to additional unobserved worker attributes. For example, it is plausible that less productive or less ambitious workers, or workers who are not prepared to spend much job effort are selfselected into part-time jobs. The results also support the property rights hypothesis as we uncover significant nonprofit wage and compensation differentials. VI. Sensitivity Analysis and Representativeness of the Data Our analysis included detailed worker and firm attributes. Examples are Organization and Important Job for workers, and Percent Subsidized, Center Age, Part-week, Extended Care, Head Start, Before and After School, Summer Camp, Evening Care, Sick Care, and 24 Hour Care for firms. To investigate the importance of controlling for these worker and firm characteristics, we omitted these variables from the system and re-estimated the models. Omission of these variables implies 44 restrictions on the unrestricted models. The calculated likelihood ratio was for the wage model, and for compensation, strongly rejecting the hypothesis that these worker and firm attributes are unimportant. Furthermore, with the omission of these variables both wage and compensation models provided different results regarding the mark-ups in comparison to the ones reported in Table 7. More specifically, the nonprofit wage mark-up for full-time workers became -5.1 percent (as opposed to the positive 5.8 percent reported in Table 7). Similarly, the nonprofit compensation differential for full-time workers (reported as 7.8 percent in Table 7) is estimated as 2.1 percent in the restricted model. In addition to the omitted variables mentioned above, we further reduced the models by omitting the following firm characteristics: On-site, National Chain, Publicly Supported, Publicly Owned, Publicly Regulated, and Religious. In other words, we made our models similar to other papers in the literature which included information on nonprofit status of the firm, but did not have additional information about the details of the 14

18 ownership. These additional omitted variables generate 16 further restrictions. We found that these additional restrictions hurt the models even further. The calculated likelihood ratio was for the wage system (while the one-percent critical value is 32 for 16 degrees of freedom), and 48.4 for the compensation system. The nonprofit wage differential for full-time workers was 4.7 percent, and the full-time mark-up in the for-profit sector was 6.2 percent. Once again, the results are different from the ones obtained from the original models. This exercise underscores the importance of detailed worker and firm characteristics, and highlights the incorrect inferences that can be made in their absence. To investigate the sensitivity of the results to variations in empirical strategy, we performed two exercises. First, we estimated the models with different number of support points. The results reported in Tables 4 and 6 are based on 5 points of support of the unobserved common factors ( ) in Equation (13). We estimated both wage and compensation equations with 5 and 4, 4 and 5, 4 and 4, as well as 6 and 6 points of support. The estimated coefficients were unaffected. Furthermore, the calculated wage and compensation mark-ups were very similar to those reported in Table 7 (based on 5 points of support). 12 Second, we investigated the sensitivity of the results to the method of modeling selection. As an alternative to the discrete factor method, we used a double-selection procedure (e.g. Main and Reilly 1992, Krishnan 1990, Lee 1982 ). In this procedure, the two selection equations are estimated using a bivariate probit, which provides estimates for selection into full-time and for-profit sector. These selection terms for each individual are entered into the four wage (or compensation) equations as additional regressors. Selection adjusted predicted wage and compensations are calculated for each individual; and their averages are reported in Table 7 in rows labeled double selection. The results obtained from this two-step procedure are consistent with those obtained from full-information maximum-likelihood, although the markups obtained from the double-selection method are somewhat larger in absolute value. Can these results be attributable to some anomaly of the child care industry? To entertain this question, we used the Five-Percent Public Use Sample (PUMS) of the 1990 census. We extracted all workers with positive wages and salary income who worked in hospitals, nursing and personal care facilities, elementary and secondary schools, colleges and universities, educational services, bus service and urban transit, and research, development and 12 Increasing the points of support beyond 6,6 imposes computational burden, and generates instability of the system. 15

19 testing services. These industries have bigger nonprofit presence than others. To investigate the raw nonprofit and full-time wage differentials in these industries we ran regressions of the logarithm of wages on nonprofit and full-time dummies as well as state dummies. The results are reported in Table 9. The nonprofit wage premium is positive in all cases, although not significant in educational services. Raw full-time wage mark-up is zero in hospitals, and negative in nursing and personal care, but positive everywhere else. Thus, the descriptive information from the 1990 census shows that the raw nonprofit and full-time mark-ups of the child care industry are similar to those observed in many other industries with a significant nonprofit presence, indicating that our results are not likely to be driven by some idiosyncratic structure of the child care industry. VI. Conclusion This paper uses a rich employer-employee matched data set to investigate the existence and the extent of nonprofit and part-time wage and compensation differentials. Utilizing data exclusively on child care workers we avoid potential contamination of results due to interindustry unobservables. The raw data reveal wage and compensation premiums in favor of nonprofit workers. They also reveal the existence of wage and compensation premiums in favor of full-time workers in for-profit centers, and compensation premium in favor of fulltime workers in the nonprofit sector. An analysis of the 1990 Census data shows that these raw differentials are similar to those found in many industries with significant nonprofit sector presence. The empirical strategy adjusts for workers self-selection into the for-profit or nonprofit sectors, as well as into full-time or part-time work. We also control for unobserved worker heterogeneity using a discrete factor model. The wage and compensation equations are estimated jointly with the selection equations using full-information maximum likelihood. We find differences between the regimes (full-time for-profit, full-time nonprofit, parttime for-profit, part-time nonprofit) in the way in which human capital of the workers are rewarded. For example, in both for-profit and nonprofit sectors, the return to education is higher for part-time workers. For full-time workers, the return to tenure is higher in for-profit centers. In nonprofit centers, the return to experience in child care for full-time workers is higher than that of part-time workers, and in for-profit centers the return to experience outside 16

20 child care is higher for full-time workers. There is substantial variation in wages as a function of employee characteristics, and there is variation in wages within sectors. For example, unionization increases wages and compensation percent, and centers that are part of a national chain provide lower wages and compensation. Other examples are the job title of the worker and the age group of the children he/she is serving. Similarly, center characteristics such as the age and the size of the center, whether the center is publicly supported, publicly owned or regulated, whether it is a religious center impact wages and compensation. Along the same lines, various programs offered by the centers (e.g. sick care, evening care) have significant impacts. After estimating the models, we calculate the wage and compensation that each worker in our sample would have received under each regime after controlling for non-random selection and unobserved worker heterogeneity. We find that potential wages of part-time workers are equivalent to those of full-time workers in the for-profit sector, and they are higher in the nonprofit sector. Part-time workers compensation is higher than that of full-timers in both sectors, and the compensation mark-up is larger than the wage mark-up. This suggests that parttime workers receive more benefits in both sectors. We find evidence of positive nonprofit wage differentials for both full-time and part-time workers. Compensation is also higher in the nonprofit sector for both types of workers. The magnitudes of these differentials are substantial. For example, nonprofit wage differential is almost six percent for full-time workers and 20 percent for part-time workers. Nonprofit compensation differential is eight percent for full-time workers and 10 percent for parttime workers. Similarly, part-time workers earn wages that are 13 percent more than full-time workers in nonprofit centers. Part-time compensation is 25 percent more than full-time compensation in nonprofit centers, and 23 percent more in for-profit centers. The same basic results are obtained when selection is modeled using a two-step procedure. These results underscore important points. First, it seems critical to control for the impact of worker self-selection into sector and into full-time work. Second, it is important to have detailed controls for both employer and employee characteristics as they have significant impacts. The results indicate that part-time jobs are good jobs in center-based child care, and they show the existence of nonprofit wage and compensation premiums, which supports the property rights hypothesis. 17

21 Table 1. Raw Full-time and Nonprofit Wage and Compensation Differentials Full-time Differential In nonprofit centers In for-profit centers W ft = W pt C ft > C pt W ft > W pt C ft > C pt (8.9%) (5.2%) (9.8%) Nonprofit Differential For full-time workers For part-time workers W np > W pr C np > C pr W np > W pr C np > C pr (7.8%) (11.6%) (11.6%) (13.6%) W stands for wage, C stands for compensation, ft represents full-time, pt represents part-time, np means nonprofit, and pr is for-profit. 18

22 Variable Table 2. Descriptive Statistics of Worker Characteristics Definition For-profit Workers Wage The hourly pre-tax wage rate * (2.023) Compensation The hourly compensation 7.418* (2.392) For-Profit Full Time Female Age Tenure Dummy (=1) if the worker works in a for-profit center Dummy (=1) if the worker works fulltime Dummy(=1) if female Age of the worker Tenure at the center (in months) Experience Years of experience in early education and child care (>10 hours per week) Other Experience Years of total work experience outside early education and child care. Less Than High School High School Some College Community College College (0.000) 0.681* (0.467) (0.148) * (10.96) * (40.751) 7.226* (6.299) (6.592) Dummy(=1) if no high school diploma 0.016* (0.126) Dummy(=1) if high school graduate or GED Dummy(=1) if some college courses (0.393) (0.471) Dummy(=1) if two year college degree (0.338) 0.232* Dummy(=1) if four year college degree (0.423) Some Graduate School Dummy(=1) if some graduate school 0.053* (0.223) Graduate Degree Dummy(=1) if graduate degree (0.211) No Training Dummy(=1) if no special training in early childhood education (0.349) Assistant Teacher Dummy(=1) if assistant teacher 0.386* (0.487) Infant-Toddler Room Own Kids Dummy(=1) if the worker works in an infant-toddler room Number of children living with worker 0.410* (0.492) 0.857* (1.221) Nonprofit Workers 7.303* (2.959) 8.567* (3.709) (0.000) 0.615* (0.487) (0.198) * (11.977) * (53.82) 8.184* (6.649) (6.630) 0.035* (0.184) (0.387) (0.476) (0.359) 0.150* (0.357) 0.083* (0.277) (0.206) (0.344) 0.461* (0.499) 0.243* (0.429) 1.102* (1.254) Full-time Workers 6.798* (2.462) (3.152) 0.504* (0.500) (0.000) (0.175) * (11.124) * (45.311) (6.346) (6.500) (0.138) (0.391) (0.474) (0.357) (0.397) (0.245) (0.197) (0.343) 0.360* (0.480) 0.357* (0.480) 0.892* (1.178) Part-time Workers 7.182* (2.775) (3.289) 0.432* (0.496) (0.000) (0.178) * (12.348) * (53.861) (6.776) (6.804) (0.192) (0.387) (0.474) (0.335) (0.383) (0.266) (0.228) (0.352) 0.544* (0.499) 0.260* (0.439) 1.153* (1.342) 19